Web caching: locality of references revisited

An effective caching strategy depends on utilizing the locality of reference. However, knowledge gained from research in traditional caches cannot be applied verbatim to web caches. We propose that web accesses exhibit at least four dimensions of locality. A logistic regression model is used to confirm our hypothesis. Based on these findings we implemented a web-caching agent to acquire knowledge about the objects it encounters, and deduce a cache strategy dynamically. We also showed that the same analysis can be applied to pre-fetching. Results on trace driven simulations of six web and proxy servers confirm that more effective caching can be achieved.